Re:Cantonese Font with Pronunciation
by nextaccountic on Tuesday May 09, 2023 @02:49AM
(#35870962)
This article is from 2010, and it says that as of 2010, progress on speech recognition flatlined since 2001
Is this still the case?
How does 2001 software compares to 2023 software? Or 2010 to 2023 for that matter
Speech recognition is leaps and bounds ahead of where it was in 2001/2010, driven by a combination of deep learning and massive datasets. I'm actually working on a similar concept (speech recognition for Mandarin language learning) and even accented speech is becoming less of a problem.
I haven't tried differentiating tones for discrete words with the recent batch of generic models but I think it might still struggle. I suspect a fine-tuned task-specific classifier might be needed for decent results.
That said, I still think your post raises some excellent points. To refine it, how about an online learning app that plays a sound or video clip from a national news TV or radio program. (Usually, the speakers have perfect pronunciation.) You repeat the words; the online app records it; then, it shows your tones/pitch/accent versus the native speaker. I think that could be incredibly useful. My point: Speech reco is an impossibly hard problem (currently) due to infinitely broad context in spoken languages. My idea would have "perfect context", so speech reco could really work.
When I was learning tones for a while, I used to record myself, then replay it. It's amazing to hear the difference between what you think you sound like and what you really sound like. An online app could help to fine tune your pronunciation very quickly.